Skip to content

Latest commit

 

History

History
87 lines (61 loc) · 3.38 KB

README.md

File metadata and controls

87 lines (61 loc) · 3.38 KB
Raw

Lucene CIFF import

Application for importing a CIFF index in Lucene (and Anserini)

Preparations:

Install Anserini

Install ciff tools

Set paths as follows:

$ export PATH=$PATH:$HOME/Workspace/anserini/target/appassembler/bin
$ export PATH=$PATH:$HOME/Workspace/ciff/target/appassembler/bin

Usage:

1. Make an index in CIFF, e.g. by using Anserini

$ IndexCollection -collection TrecCollection -index anseriniIndex \
    -input src/test/resources/test_docs/ -optimize -threads 1 \
    -stemmer none -keepStopwords

$ IndexReaderUtils -index anseriniIndex/ -stats

$ SearchCollection -index anseriniIndex -topicreader Trec \
    -stemmer none -keepstopwords -qljm -output anseriniIndex.run \
    -topics src/test/resources/test_topics.txt

$ ExportAnseriniLuceneIndex -output anseriniIndex.ciff \
    -index anseriniIndex -description "Test description"

2. Hook into Lucene to turn the CIFF file into a Lucene index

To generate a Lucene index from CIFF files, we make use of Lucene's codecs module, which offers an API for reading and writing parts of the index in different codecs. We create wrappers around the 'reading' part of the codecs module, that read from CIFF and offer the data in the correct format (such as TermsEnum, Fields, DocValues). We then use the 'writing' part of the codecs module to write the data into an index that Lucene can read later on.

Because we convert the CIFF data to the internal Lucene formats, the import is codec agnostic. In other words, we can write the index in different codecs, such as the SimpleText codec for easy debugging, or the Lucene90 codec for efficient retrieval.

To run the application, we can issue the following command:

$ mvn compile exec:java \
    -Dexec.mainClass="nl.ru.ciffimporter.CiffImporter" \
    -Dexec.args="INPUT OUTPUT [CODEC]"

Alternatively, we can package the importer into a standalone application as follows:

$ mvn clean package appassembler:assemble
$ ./target/appassembler/bin/CiffImporter INPUT OUTPUT [CODEC]

INPUT should be the CIFF file you are trying to import. OUTPUT is the (non-existing) directory that will be created for the Lucene index. CODEC is an optional argument that can be used to specify the codec Lucene will use for writing out the index. For instance, you can provide SimpleText as codec to generate a human-readable text index. If not specified, the default Lucene codec (i.e. the latest version) will be used.

Sanity check: CIFF -> Lucene -> CIFF

If we import a CIFF index into Lucene, and then re-export it, we see that the result is identical to the original CIFF file.

Note: the -description argument to ExportAnseriniLuceneIndex must be identical to the description in the original CIFF file.

$ mvn compile exec:java \
    -Dexec.mainClass="nl.ru.ciffimporter.CiffImporter" \
    -Dexec.args="anseriniIndex.ciff luceneIndex"

$ ExportAnseriniLuceneIndex -output luceneIndex.ciff \
    -index luceneIndex -description "Test description"

$ diff -s anseriniIndex.ciff luceneIndex.ciff
Files anseriniIndex.ciff and luceneIndex.ciff are identical